Author: GuyHaworth
Date: 23:41:54 01/29/04
Go up one level in this thread
There seem to be 3 themes intertwined here: - reducing the index-range of the uncompressed EGT - compressing the EGT in the best way - accessing the compressed EGT Although the last two points affect each other, the first can be taken separately. Eugene does not avoid giving an index-position to every unreachable position. I suspect that even 8/8/8/8/8/8/8/K1k1R3 w has an index position, even though the sntmK is in check. What Eugene avoids is positions with 'unblockable checks', e.g. 8/8/8/8/8/8/8/K1kR4 w. The ICGA ACG9 Conference Proceedings (book) and also v23-3 of the ICGA_Journal tabulated how many 'broken' positions - i.e. with sntmK in check or two pieces attempting to occupy the same square - are left by this technique. Thus, since all checks by Ns are unblockable, there are no broken positions in, e.g. KNNNKN's EGT or any half-EGT where the stm has no non-Ns. At the other extreme, 62.94% of wtm KQQQK positions are broken re bK-in-check. As for the compression, this is by Andrew Kadatch and you can look up his work on the web. Good compression techniques came out of the FSU, maybe because their computers were smaller. Compression technique is, to some extent, shaped to the data being compressed and the subsequent intended use. EN compresses 8KByte-blocks so that 'only' this much EGT has to be decompressed to find the actual position you seek. However, John Tamplin and I note that for our EdZ50Z EGTs, where we store the difference between DTZ50 and DTZ (v often zero), we can have a string of 8KB blocks which are all zeros. In the extreme case, KNPKQ DTZ50 has just one position which was a DTZ win and is now a draw. There are no 'delayed wins' with DTZ50 > DTZ. Thus, all 8KB blocks are 'null' except 1, which has just one non-zero 'draw' entry. So that's an example of where we would be happy to vary the compression scheme - by having a larger block size. Although AK's scheme can be 'finessed' [e.g. by choosing compression scheme at block level], it is vg as EN says, and the gains appear not to be major at the moment. g
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.